Perplexity Sonar Online: The Search-Augmented LLM API for Real-Time Data
Perplexity's Sonar API returns LLM-generated answers with inline citations from live web search - an OpenAI-compatible endpoint that replaces custom RAG pipelines for real-time data retrieval use cases.
Standard LLM APIs return text generated from training data with a knowledge cutoff. Perplexity Sonar returns answers grounded in live web search results, with citations in the response body. The model is not just searching and summarizing - it integrates search results into coherent prose with numbered references.
This replaces a common pattern: LLM + search tool + citation extraction + response formatting. Sonar does all of it in one API call.
Sonar vs Sonar Pro
Sonar
Sonar Pro
Underlying LLM
Smaller Sonar model
Larger Sonar model
Search sources
Standard index
Expanded sources, more recent
Price
$1/1000 searches + $1/1M tokens
$5/1000 searches + $5/1M tokens
Citations
Yes
Yes, more thorough
Context
127k
127k
For most use cases, Sonar is sufficient. Use Sonar Pro when you need maximum coverage on a fast-moving topic or when citation accuracy is critical (legal research, medical information, financial news).
Team workspace
Ship faster with chat, meetings, and projects in one place — Zlyqor.
from openai import OpenAI
# Drop-in swap from OpenAI
client = OpenAI(
api_key=os.environ["PERPLEXITY_API_KEY"],
base_url="https://api.perplexity.ai",
)
response = client.chat.completions.create(
model="sonar",
messages=[
{
"role": "system",
"content": "You are a research assistant. Cite your sources.",
},
{
"role": "user",
"content": "What are the latest benchmark results for GPT-4.1 vs Claude 3.7?",
},
],
)
answer = response.choices[0].message.content
# Citations are available in response.citations (Perplexity extension)
citations = getattr(response, "citations", [])
print(answer)
for i, citation in enumerate(citations, 1):
print(f"[{i}] {citation}")
The search_recency_filter Parameter
For time-sensitive queries, the search_recency_filter parameter constrains Sonar to only return sources from a specific time window:
response = client.chat.completions.create(
model="sonar",
messages=[
{"role": "user", "content": "What AI models were released this week?"},
],
extra_body={
"search_recency_filter": "week", # "month", "week", "day", "hour"
"return_images": False,
"return_related_questions": True,
},
)
Setting search_recency_filter: "day" ensures you only get information from the past 24 hours - critical for market monitoring, breaking news summarization, or competitive intelligence.
Use Cases
Competitive analysis: "Summarize the key announcements from [competitor] in the last 30 days" - run this on a schedule and store results for trend tracking.
News monitoring: "What are the top stories about AI regulation this week?" - replaces a custom pipeline of news API + LLM summarization + deduplication.
Live data extraction: "What is the current valuation of [company] according to recent news?" - works where data is too recent for training data.
Research grounding: When building a RAG system on proprietary documents, use Sonar to supplement with real-time web context that your documents may not cover.
When to Use Sonar vs Standard RAG
Use Sonar when:
Data changes frequently (news, prices, regulations, product releases)
You do not have a corpus to index
Setup time matters (Sonar is ready in one API call, RAG requires indexing infrastructure)
Use standard RAG when:
Data is proprietary and cannot be sent to a third-party search index
You need exact retrieval from specific documents
Volume is high enough that $5/1000 searches is cost-prohibitive
You need strict citation to specific internal documents
Practical deep-dives on LLMs, developer tools, and AI engineering. No filler. Unsubscribe any time.
// written byFIG. AUTH-01
530
Mahmudul Haque Qudrati
CEO & ML Engineer
CEO and ML Engineer at Pristren. Builds AI-powered software for teams and writes about machine learning, LLMs, developer tools, and practical AI applications.
What is Harness engineering: Leveraging Codex in an agent-first world? A Practical Overview
Harness engineering is the practice of building structured, safe environments for AI agents to execute code. This post explains how to leverage OpenAI Codex in an agent-first world, with concrete examples, cost breakdowns, and honest tradeoffs.
